Re: [PATCH] ata: libata-core: Enable context analysis

[Niklas Cassel <cassel@kernel.org>]

Hello Bart,

On Tue, May 12, 2026 at 01:00:58PM -0700, Bart Van Assche wrote:
> On 5/11/26 9:53 AM, Niklas Cassel wrote:
> > 1) It seems that you are only checking for the EH mutex.
> > 
> > $ git grep -A 1 "LOCKING" drivers/ata/
> > 
> > Does have many functions with:
> > 	LOCKING:
> > 	spin_lock_irqsave(host lock)
> > 
> > Would it be possible to add __must_hold() annotations for these functions
> > too, but for ap->lock instead of EH mutex ?
> 
> This is something I can't do myself. This is something that should be done
> by an ATA expert. I'm not an ATA expert.

In my mind you are as much of an ATA expert as me and Damien :)


> 
> > 2) There seems to be some files that did not get any annotations, e.g.
> > drivers/ata/libata-scsi.c, drivers/ata/libata-sata.c,
> > drivers/ata/libata-acpi.c, drivers/ata/libata-pmp.c.
> > 
> > Would it be possible to add annotations for these files too?
> 
> Just like for (1), since there are inconsistencies between the
> "LOCKING:" documentation and the implementation, this is best done by
> an ATA expert.

Are there really a lot of inconsistencies though?

If you simply add annotations to e.g. drivers/ata/libata-scsi.c and
drivers/ata/libata-sata.c, which matches the kdoc, how many functions are
there that will not compile?

Considering how old and mostly unchanged most of these functions are, I
would expect the vast majority to actually be consistent with the kdoc.


> 
> The goal of the annotations in this patch is to make sure that the build
> doesn't break with CONTEXT_ANALYSIS := y. Any additional annotations can
> be implemented as follow-up patches.

Sure, doing so in follow-up patches would also be fine, although,
personally I would prefer to include at least a few ap->lock annotations
in this patch, so that people can see that it is possible to have
annotations also for ap->lock and not only EH mutex.


I could see in another thread that function pointer annotations require
clang 23:
https://lore.kernel.org/all/acqELlVC6vEeEF-W@elver.google.com/

Do you want us to pick up the patch in $subject now, or wait for the above
to land first?


Oh, there were some Sashiko review comments on the patch in $subject:
https://sashiko.dev/#/patchset/20260505042227.909666-1-bvanassche%40acm.org

Could you please check if they are valid review comments?


Kind regards,
Niklas